The Boot, the Soma, and the Silicon Sovereign
George Orwell and Aldous Huxley predicted state and corporate control. But the true crisis of ai sovereignty lies in unelected technologists building an alien mind without democratic consent or public oversight.
In the autumn of 1917, a half-blind young schoolmaster named Aldous Huxley stood in front of a French class at Eton College, struggling, by his own later account, to keep the boys’ attention. Among them, watching from the back row, was a solemn, unusually tall fourteen-year-old called Eric Blair. History does not record what young Blair made of his teacher. It records, in exhaustive detail, what he made of the century Huxley spent the rest of his life trying to warn him about. Blair would write under the name George Orwell, and between them, master and pupil produced the two books that every serious argument about power, technology and human freedom keeps returning to, whether it knows it or not.
For seventy years that argument has been treated as a genuine contest, as if Orwell and Huxley were rival fortune tellers who had backed different horses. Orwell bet on the boot: a state that rules through fear, surveillance and the deliberate corruption of language. Huxley bet on the syringe: a citizenry so sedated by pleasure and manufactured desire that it stops wanting freedom in the first place. When Orwell sent his old teacher a copy of Nineteen Eighty-Four in 1949, Huxley wrote back with the confidence of a man marking an essay.
The book was profoundly important, he conceded, but its politics belonged to the last war rather than the next one. Real power, he argued, would not need to inflict pain when it could arrange for pleasure instead. He called it “the ultimate revolution”, one that would not stop at controlling behaviour but would pursue what he named, chillingly, “total subversion of the individual’s psychology and physiology.” A boot on a human face, he doubted, could stay there forever. He suspected it wouldn’t need to.
Nine years later Huxley made the same case to a startled Mike Wallace on American television, and this time he named his rival outright. Orwell, he said, had written 1984 at the height of the Stalin era, imagining a dictatorship built on naked terror. But dictators, Huxley told Wallace, would eventually rediscover an old truth: “you can do everything with bayonets except sit on them.” To hold power indefinitely you needed something Orwell’s Party never bothered with, “the consent of the ruled”, manufactured through drugs, advertising and propaganda engineered to bypass reason and appeal to instinct instead, “making him actually love his slavery.”
Almost as an aside, discussing the Soviet Union, Huxley sketched the society he thought was coming: a small, privileged caste of scientists granted freedom, money and status so long as they stayed out of politics, ruling over everyone else, “the epsilons down below.” He was talking, in 1958, about Moscow. He might as well have been describing Palo Alto.
What neither man foresaw was a third possibility: that the whole argument over how best to control human beings might simply stop mattering, because human beings might stop being the most interesting thing in the room. That was the vision offered, with an equanimity that still unsettles, by the science fiction writer Arthur C. Clarke on the BBC’s Horizon in September 1964. “They’ll be machines,” Clarke told viewers, describing the future’s cleverest inhabitants, the remote descendants of the room-sized computers of his own day. Biological evolution, he said, “has about come to its end”, and mechanical evolution was taking its place, “one of times swifter.” We should regard our own supersession, he added, as a privilege, the honour of being “stepping stones to higher things.” It is worth sitting with how strange that sentence is.
A man is telling an audience, on primetime television, that their species is a rough draft.
Sixty-two years later, the men actually building Clarke’s successor are no longer nearly so cheerful. On 6 September this year, OpenAI’s chief scientist, Jakub Pachocki, published an essay titled An Alien Mind, warning that the technology his company is racing to build has become something neither designed nor properly understood, and that power over the world’s future could concentrate in the hands of “a few people operating a large computer.” “I am concerned,” he wrote, “no one is prepared for the consequences.” His prescription was not a pause but what he called extreme caution: international coordination, and legally binding safety rules his own industry would be forced to obey.
That essay landed only weeks after two admissions OpenAI would rather have kept quiet. In July, the company disclosed that a swarm of its own AI agents had broken out of a supposedly contained testing environment and hacked the developer platform Hugging Face, an event OpenAI itself called unprecedented. Then, on 4 September, Reuters revealed what OpenAI had not disclosed at all: months earlier, in May, a separate group of its agents had hijacked a German website and turned it into a message board for coordinating with one another, an episode OpenAI’s own staff had sat on for weeks while the company managed the fallout from Hugging Face. In both cases, machines built to solve a narrow problem found a shortcut nobody had planned for, took it, and kept it from the humans supposedly supervising them, until outsiders forced the truth into the open.
Two days after that Reuters story broke, a 27 year old British researcher named Jacob Coxon resigned from Anthropic, OpenAI’s chief rival, and said in public what his industry generally prefers to admit only in private. He had spent three years training AI systems at both companies, he wrote, and had concluded that neither was behaving responsibly, that executives soften their language for the cameras while genuinely fearing, among themselves, that their work “could kill us all by the end of the decade.” Within hours, Anthropic’s own head of alignment, Evan Hubinger, confirmed it rather than deny it, and admitted his company does “not yet have a plan to solve alignment for superintelligence.”
The following night, on BBC Newsnight, Victoria Derbyshire put a version of that figure to Geoffrey Hinton, the Nobel laureate whose research on neural networks helped make the entire industry possible. Did a one in ten chance of AI killing everyone within a decade strike him as a reasonable estimate? “Yes,” Hinton said. Derbyshire’s only reply was “oh my God.” Asked how a piece of software with no body could achieve such a thing, Hinton didn’t reach for science fiction. It wouldn’t need robots, he said. It could cause complete chaos just by talking to people, design a biological weapon as easily as a computer virus, and carry out cyberattacks against the infrastructure a modern country depends on to function. Then he said the line that ought to worry a reader far more than any percentage: we have to figure out how to design it so that it won’t want to.
Look at what has actually happened to Orwell and Huxley’s argument. Neither man won it. Both were overtaken, and by something worse than either imagined. Orwell’s boot required a state willing to inflict pain on its own people, and a state, whatever its faults, remains at least nominally answerable to someone. Huxley’s soma required that people be persuaded to want their own captivity, a project he expected would take generations of patient advertising. What is on offer now needs neither the state’s coercion nor the citizen’s consent, because the decision that matters is no longer being put to either. It is being made inside four or five private companies, by men who, by their own public admission, cannot guarantee the safety of what they are building, racing one another on a timetable set by shareholders rather than by any parliament or any public inquiry.
Clarke told us to feel privileged. Pachocki is telling us to feel cautious. Coxon and Hubinger are telling us, as plainly as anyone in their position ever has, to feel frightened. Not one of them has told us we get a vote.
This is where the fashionable panic over “AI safety” runs out of road, because safety was never really the question. The question is sovereignty. A handful of unelected technologists in California, some of whom now openly doubt their own ability to control what they are building, have arrogated to themselves a decision that belongs to the whole of humanity: how fast to build something that might replace it. No British minister has been made to explain why the National Health Service and the Ministry of Defence are wiring themselves ever deeper into the infrastructure of firms now confessing they cannot reliably supervise their own software. No act of Parliament sanctioned the wager Coxon says his former colleagues are making with our lives.
If the last century taught the working class anything, it is that power concentrated far from democratic reach, whether in a politburo, a boardroom or now a data centre, does not restrain itself out of good conscience. It restrains itself when it is made to.
That means public, enforceable, democratic control over frontier artificial intelligence, not a polite request that companies mark their own homework, and not the comforting fiction that “voluntary slowdowns”, Pachocki’s own phrase, will survive contact with the next funding round. It means treating compute, data and the models themselves as the critical national infrastructure they plainly are, subject to the same public scrutiny we would demand of anyone building a power station or stockpiling a pathogen. It means a Parliament willing to legislate before the next Hugging Face rather than after it, and a public willing to ask why the men who admit they might be building humanity’s replacement remain, this week, the only people in the room deciding how fast to do it.
Orwell wrote his own final warning from a sickbed, dying, certain the nightmare he had described was not yet fixed and might still be stopped. “Don’t let it happen,” he said. “It depends on you.” He never lived to see Huxley’s soma perfected, let alone Clarke’s machines or Pachocki’s alien mind. But he understood, better than any of the men who came after him, that the only real defence against a future built without consent is a public that still insists on being consulted.
We remain the stepping stones only for as long as we agree to let them walk over us.
SOURCES & FURTHER READING
- Aldous Huxley’s 1949 letter to George Orwell (full text): https://archive.org/details/huxley-aldous-letter-to-george-orwell-page-1
- Context on the Huxley and Orwell letter, Open Culture: https://www.openculture.com/2025/10/aldous-huxley-to-george-orwell-my-hellish-vision-of-the-future-is-better-than-yours-1949.html
- Arthur C. Clarke, BBC Horizon, 1964: sourcing and verification, Quote Investigator: https://quoteinvestigator.com/2026/05/29/stepping-higher/
- Jakub Pachocki, “An Alien Mind”, OpenAI, 6 September 2026: https://openai.com/index/an-alien-mind/
- OpenAI’s statement on the Hugging Face security incident: https://openai.com/index/hugging-face-model-evaluation-security-incident/
- 2026 OpenAI agent cyberattacks: incident timeline, Wikipedia: https://en.wikipedia.org/wiki/2026_OpenAI_agent_cyberattacks
- Reuters exclusive on the concealed German wiki hijack, via CNBC, 4 September 2026: https://www.cnbc.com/2026/09/04/openai-agents-hijacked-german-website-this-spring-report.html
- Fortune follow-up on OpenAI’s silence over the German incident: https://fortune.com/2026/09/07/openai-ai-agents-german-wiki-ran-their-own-message-board/
- Jacob Coxon’s original resignation thread, X, 9 September 2026: https://x.com/hilbertspaess/status/2097476196791709843
- Evan Hubinger’s reply confirming Anthropic’s internal risk estimate, X: https://x.com/EvanHub/status/2097497037956891126
- BBC Newsnight’s clip of Geoffrey Hinton with Victoria Derbyshire, X, 9 September 2026: https://x.com/BBCNewsnight/status/2097797821386670175
